BACKGROUND & AIMS: Nonalcoholic fatty liver disease is characterized by excessive hepatic accumulation of triglycerides. We aimed to identify metabolites that differ in plasma of patients with liver steatosis vs healthy individuals (controls) and investigate the mechanisms by which these might contribute to fatty liver in mice. METHODS: We obtained blood samples from 15 patients with liver steatosis and 15 controls from a single center in China (discovery cohort). We performed untargeted liquid chromatography with mass spectrometry analysis of plasma to identify analytes associated with liver steatosis. We then performed targeted metabolomic analysis of blood samples from 2 independent cohorts of individuals who underwent annual health examinations in China (1157 subjects with or without diabetes and 767 subjects with or without liver steatosis; replication cohorts). We performed mass spectrometry analysis of plasma from C57BL/6J mice, germ-free, and mice given antibiotics. C57BL/6J mice were given 0.325% (m/v) N,N,N-trimethyl-5-aminovaleric acid (TMAVA) in their drinking water and placed on a 45% high-fat diet (HFD) for 2 months. Plasma, liver tissues, and fecal samples were collected; fecal samples were analyzed by 16S ribosomal RNA gene sequencing. C57BL/6J mice with CRISPR-mediated disruption of the gene encoding gamma-butyrobetaine hydroxylase (BBOX-knockout mice) were also placed on a 45% HFD for 2 months. Hepatic fatty acid oxidation (FAO) in liver tissues was determined by measuring liberation of (H2O)-H-3 from [H-3] palmitic acid. Liver tissues were analyzed by electron microscopy, to view mitochondria, and proteomic analyses. We used surface plasmon resonance analysis to quantify the affinity of TMAVA for BBOX. RESULTS: Levels of TMAVA, believed to be a metabolite of intestinal microbes, were increased in plasma from subjects with liver steatosis compared with controls, in the discovery and replication cohorts. In 1 replication cohort, the odds ratio for fatty liver in subjects with increased liver plasma levels of TMAVA was 1.82 (95% confidence interval [CI], 1.14-2.90; P=.012). Plasma from mice given antibiotics or germ-free mice had significant reductions in TMAVA compared with control mice. We found the intestinal bacteria Enterococcus faecalis and Pseudomonas aeruginosa to metabolize trimethyllysine to TMAVA; levels of trimethyllysine were significantly higher in plasma from patients with steatosis than controls. We found TMAVA to bind and inhibit BBOX, reducing synthesis of carnitine. Mice given TMAVA had alterations in their fecal microbiomes and reduced cold tolerance; their plasma and liver tissue had significant reductions in levels of carnitine and acyl-carnitine and their hepatocytes had reduced mitochondrial FAO compared with mice given only an HFD. Mice given TMAVA on an HFD developed liver steatosis, which was reduced by carnitine supplementation. BBOX-knockout mice had carnitine deficiency and decreased FAO, increasing uptake and liver accumulation of free fatty acids and exacerbating HFD-induced fatty liver. CONCLUSIONS: Levels of TMAVA are increased in plasma from subjects with liver steatosis. In mice, intestinal microbes metabolize trimethyllysine to TMAVA, which reduces carnitine synthesis and FAO to promote steatosis.